How do you handle provider failures during audio and video generation?

How do you handle provider failures during audio and video generation?

External services can reject requests, time out, or return incomplete output. Recovery should reflect what is known about the operation’s outcome.

Separate retryable and corrective failures

A temporary service outage may justify a bounded retry with backoff. Invalid input, unsupported options, or missing credentials need correction. Keep provider errors in diagnostic records while giving users a concise explanation. Avoid exposing secrets or raw private requests in logs.

Resolve uncertain outcomes

If the provider returns a job identifier, preserve it and check status before starting another generation. A timeout does not prove that no work occurred. Use supported idempotency mechanisms where available and record attempts in your own application. Test a response lost after acceptance, not just a clean rejection. This reveals duplicate-generation risks that a simple retry loop misses.